Correlation Filter and Deep Siamese Network Hybrid Algorithm for Visual Object Tracking
Ying Min Hou, Xinyu Lin, Jiao Li · 2021
In practical applications, the tracking performance and the computational efficiency play an important role in object tracking. According to the significant computational efficiency of correlation filter tracking algorithms and the excellent tracking performance of deep learning-based tracking algorithms, in this paper we proposed Correlation Filter and deep Siamese Network hybrid algorithm for object tracking (CF\_SIAM). A video grouping object tracking strategy is designed, which uses the correlation filtering tracking algorithm within the group, and applies the tracking algorithm with deep Siamese network among groups to revise the tracking target. The experimental results show that our algorithm can significantly improve tracking performance. On the OTB100 dataset, the AUC success rate of our CF\_SIAM algorithm outperforms KCF and SiameseFC by 11.5% and 3.2% respectively.